Our thinking

AI’s hidden footprint

7 June, 2024

When you think of high-emissions industries, air travel or oil and gas probably come to mind. But what if we told you the tech industry and cloud infrastructure actually account for more global greenhouse gas emissions than commercial flights? Take large language models like ChatGPT. Your favourite timesaving application is one of the most energy-intensive digital tools around. In fact, estimates suggest that the energy needed for data centres could triple by 2030, making up 7.5% of U.S. energy consumption.

And it’s not just emissions. Water consumption is also soaring. Data centres are heavily reliant on water in evaporative cooling systems to keep equipment from overheating. Research indicates that cooling the machines that trained ChatGPT-3 at Microsoft’s data facilities consumed about 700,000 litres of water. Giants like Microsoft, Google, and Meta have all increased their water usage in recent years, as they expand their AI and cloud infrastructure. Experts predict that AI demand could drive water withdrawal to between 4.2 billion and 6.6 billion cubic meters by 2027, about half the annual water consumption of the UK.

Yet AI’s carbon and water footprint remains shrouded in mystery, with many unaware of tech’s substantial ecological impact. Increased transparency and reporting from tech companies are essential to track and fully understand the environmental costs of AI models.

While we can hope AI might one day solve this problem itself, it’s important to consider the environmental impact as you and your business lean more on AI models like ChatGPT. After all, every bit of tech we use has a footprint, so let’s make sure we’re stepping lightly.

By Meg Seckel